WO2009027889A2 - Uncertainty maps for segmentation in the presence of metal artifacts - Google Patents
Uncertainty maps for segmentation in the presence of metal artifacts Download PDFInfo
- Publication number
- WO2009027889A2 WO2009027889A2 PCT/IB2008/053236 IB2008053236W WO2009027889A2 WO 2009027889 A2 WO2009027889 A2 WO 2009027889A2 IB 2008053236 W IB2008053236 W IB 2008053236W WO 2009027889 A2 WO2009027889 A2 WO 2009027889A2
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- model
- image
- metal
- patient image
- mesh
- Prior art date
Links
- 239000002184 metal Substances 0.000 title claims abstract description 91
- 230000011218 segmentation Effects 0.000 title claims abstract description 50
- 238000000034 method Methods 0.000 claims abstract description 54
- 230000009467 reduction Effects 0.000 claims abstract description 9
- 210000000056 organ Anatomy 0.000 claims description 25
- 239000013598 vector Substances 0.000 claims description 17
- 210000003484 anatomy Anatomy 0.000 claims description 12
- 238000003709 image segmentation Methods 0.000 claims description 5
- 238000002591 computed tomography Methods 0.000 description 9
- 230000006978 adaptation Effects 0.000 description 7
- 230000008901 benefit Effects 0.000 description 4
- 238000011540 hip replacement Methods 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 4
- 239000007943 implant Substances 0.000 description 4
- 238000012937 correction Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 241000276498 Pollachius virens Species 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000002599 functional magnetic resonance imaging Methods 0.000 description 2
- 210000001624 hip Anatomy 0.000 description 2
- 238000002595 magnetic resonance imaging Methods 0.000 description 2
- 210000004197 pelvis Anatomy 0.000 description 2
- 238000002600 positron emission tomography Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 210000002307 prostate Anatomy 0.000 description 2
- 238000002603 single-photon emission computed tomography Methods 0.000 description 2
- 230000003187 abdominal effect Effects 0.000 description 1
- 238000010317 ablation therapy Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 239000006227 byproduct Substances 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013170 computed tomography imaging Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 210000002216 heart Anatomy 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 210000003734 kidney Anatomy 0.000 description 1
- 210000004185 liver Anatomy 0.000 description 1
- 210000004072 lung Anatomy 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000000116 mitigating effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000001959 radiotherapy Methods 0.000 description 1
- 210000003079 salivary gland Anatomy 0.000 description 1
- 210000004872 soft tissue Anatomy 0.000 description 1
- 238000001356 surgical procedure Methods 0.000 description 1
- 238000002560 therapeutic procedure Methods 0.000 description 1
- 210000000115 thoracic cavity Anatomy 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 210000003932 urinary bladder Anatomy 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/80—Geometric correction
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
- A61B6/5258—Devices using data or image processing specially adapted for radiation diagnosis involving detection or reduction of artifacts or noise
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/149—Segmentation; Edge detection involving deformable models, e.g. active contour models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2200/00—Indexing scheme for image data processing or generation, in general
- G06T2200/04—Indexing scheme for image data processing or generation, in general involving 3D image data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10104—Positron emission tomography [PET]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10108—Single photon emission computed tomography [SPECT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20004—Adaptive image processing
- G06T2207/20012—Locally adaptive
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30052—Implant; Prosthesis
Definitions
- the present application finds particular application in model-based segmentation systems, particularly involving adaptive subject imaging. However, it will be appreciated that the described technique may also find application in other subject imaging systems, other image adaptation scenarios, or other medical techniques.
- Metal implants such as hip replacements or dental fillings are a source of severe artifacts in CT imaging.
- a prominent and undesirable effect is that metal implants cause beam-hardening, which causes dark and bright shades close to and radiating from the metal.
- Metal artifacts impede proper diagnosis, as well as automatic or semi-automatic organ segmentation.
- Metal artifact reduction methods can improve subjective image quality for the observer. Regardless of the visual improvement, these methods are often based on removal of the artifacted voxels and replacement by an interpolation of nearby voxels, which introduces an additional uncertainty, such that that there is no guarantee that original structures are recovered by the method.
- structures are often unaccounted for after metal artifact reduction, and other structures may be undesirably introduced. This imposes another impediment for automatic organ segmentation.
- Image reconstruction and/or segmentation can be detrimentally affected by metal in a patient.
- model-based reconstruction a mesh model of an organ is superimposed on a diagnostic image or diagnostic data. Two opposing forces, internal energy through which the model tries to hold its shape, and external energy that urges the model towards the interfaces or surfaces in the diagnostic image, work concurrently in opposition to bring the model to the shape of the organ. While model- based segmentation is satisfactory when the edges of the organ are well-defined in the image, problems can arise when there is artifact degradation, such as the metal artifacts discussed above.
- metal objects e.g., pins, screws, hip replacements, filings, etc.
- artifacts such as streaks, which obscure image detail, increasing uncertainty regarding the exact location of the surface of the organ.
- present application provides new and improved model-based segmentation systems and methods that account for uncertainty associated with image data obscured by metal artifacts, which overcome the above-referenced problems and others.
- a system for system for image segmentation in the presence of metal artifacts includes a model generator that receives patient image data and stores trained models of anatomical structures, and a voxel analyzer that determines whether metal artifacts are present in one or more voxels in the patient image data.
- the system further includes a processor that executes a metal artifact reduction algorithm and generates an uncertainty map with corrected voxel data incorporated therein for a patient image generated from the patient image data, and a segmentation tool that conforms a trained model of an anatomical structure corresponding to the patient image and segments the patient image using a model-based segmentation technique and evaluates the uncertainty derived by the processor.
- a method for performing model-based segmentation in includes generating a patient image of a region of a patient that includes a metal object, generating an uncertainty map for the patient image indicative of uncertainty in portions of the patient image due to metal object reconstruction artifacts, and employing the uncertainty map when segmenting a portion of the patient image displaced from the metal object using model-based segmenting.
- the apparatus further includes means for generating an uncertainty map with interpolated data that reduces the metal artifacts, means for weighting features associated with regions of a surface of a mesh model, and means for segmenting the 3D patient image using model- base segmentation to conform the mesh model to the 3D patient image.
- a method of performing model-based segmentation comprises generating a diagnostic image of an organ or region of a subject, which image has metal artifacts, selecting a model of the imaged organ or region, and applying a first force on the model, which urges the model to hold its initial shape.
- the method further comprises applying a second force on the model, which urges the model to deform into alignment with corresponding interfaces in the diagnostic image, and, on an image voxel by image voxel basis, adjusting the first and second forces in accordance with the degree of metal artifacting, such that the magnitude of the first force relative to the second force increases for voxels with a higher degree of metal artifacting, and the magnitude of the second force relative to the first force increases for voxels with a lower degree of metal artifacting.
- Yet another aspect relates to a system for image segmentation in the presence of metal artifacts, including a model generator that receives patient image data and stores trained models of anatomical structures, a voxel analyzer that determines whether metal artifacts are present in one or more voxels in the patient image data, and a processor that generates an uncertainty map without changing the patient image data.
- the system further includes a segmentation tool that conforms a trained model of an anatomical structure corresponding to the patient image and segments the patient image using a model- based segmentation technique and evaluates the uncertainty derived by the processor.
- One advantage is that image segmentation accuracy is improved.
- Another advantage resides in mitigating metal artifact distortion in segmented images.
- Another advantage resides in reducing total energy expended to deform a model to an image, which improves processing speed.
- FIGURE 1 illustrates a system that adjusts the external energy applied to deform an image in accordance with the likelihood that a voxel in the image is contaminated by metal artifacts.
- FIGURE 2 illustrates an image of a hip with two implants, and an uncertainty map such as may be generated by the voxel analyzer.
- FIGURE 3 shows an example of the technique as applied to the example of Figure 2, together with a metal artifact reduction method
- FIGURE 4 illustrates a diagram of the basic principles of model-based segmentation.
- FIGURE 5 is an example of segmented image adaptation of the image and a trained model using external forces that push inward on the model to adapt the model to the image.
- FIGURE 6 shows features of the model and their associations with features of the image, which delineate the internal energies of the respective features.
- FIGURE 7 illustrates an adaptation iteration wherein the mesh model is adapted to the image. It is to be appreciated that model-based segmentation is an iterative algorithm.
- FIGURE 8 illustrates an example of an internal force instruction in execution, such as may be executed or prompted by the segmentation tool in conjunction with a CT scanner or the like.
- FIGURE 9 is an illustration of a screenshot showing an organ surface with a mesh model surface overlaid thereon.
- FIGURE 10 illustrates a method for generating uncertainty maps in to mitigate distortion caused by metal objects in a patient when segmenting a nuclear or radiographic image of the patient.
- FIGURE 11 illustrates a method of generating an uncertainty map to adapt a model of an anatomical structure for segmenting a patient image.
- the systems and methods described herein facilitate segmenting an image of soft tissue whose surface boundaries are artifacted and rendered uncertain by artifacts from a metal object in other regions of an imaged subject.
- a metal ball in a hip replacement can cause a starburst-like artifact across an image of a patient's pelvic region, partially obscuring organs of interest (e.g., bladder, prostate, etc.).
- organs of interest e.g., bladder, prostate, etc.
- the artifact obscures the precise position of the metal itself.
- others have proposed trying to identify the exact location of the metal object, using this information to estimate a correction to the unreconstructed data, and reconstructing the corrected data.
- the remote organs are segmented by overlaying a mesh model having the shape of a normal organ. Opposing forces are applied to the mesh model. One force is applied to move the surface of the mesh model into alignment with surface boundaries in the image.
- the other force is applied to hold the mesh in the nominal organ shape.
- the relative magnitudes of these forces are adjusted in accordance with the uncertainty. It will be appreciated that this embodiment is applicable to uncertainties due to other factors, and does not require performing a metal artifact correction technique as a preliminary step.
- the system includes a subject imaging device (SID) 12 that is coupled to an image data memory 13 and a reconstruction processor 14, which receives captured data from the SID and reconstructs the data into an image.
- the SID can be, for instance, a computed tomography (CT) scanner, a positron emission tomography (PET) scanner, a single photon emission computed tomography (SPECT) scanner, a magnetic resonance imaging (MRI) device, a functional MRI (fMRI) device, or some other suitable subject imaging device.
- CT computed tomography
- PET positron emission tomography
- SPECT single photon emission computed tomography
- MRI magnetic resonance imaging
- fMRI functional MRI
- the system further includes a model generator 16, which comprises a processor 18 and memory 20.
- the memory can store information associated with one or more models of anatomical structures, which may be generated from one or more clinical images of subjects.
- the memory can store 3D models of individual organs or other structures (e.g., heart, liver, kidney, bladder, prostate, lung, salivary gland, etc.), as well as groups of organs and/or structures (e.g., liver-and-kidney, heart-and-lung(s), pelvic region, abdominal region, thoracic region, cranial region, etc.).
- the stored models are models of the organ or structure for a nominal person and may further include models for the inter-patient variability of these organs.
- the model generator further includes a voxel analyzer 22 that quantitatively and qualitatively analyzes a subject image for artifacts caused by metal in the subject (e.g., fillings, hip replacements, pins, screws, etc.), and a weighting module 24 that weights the external energy locally at a voxel position according to the likelihood metal artifacts being present at this location.
- the model generator includes a segmentation tool 26 that applies external and internal forces to a selected model to conform the model to the subject image for segmentation while accounting for artifacts caused by metal objects.
- the system comprises a graphical user interface (GUI) 28, which presents image and/or model information to a user and into which a user enters information to manipulate images, models, etc.
- GUI graphical user interface
- a metal identification protocol is applied by the processor 18, in conjunction with the SID 12, to identify the location of the metal.
- the voxel analyzer 22 qualitatively predicts whether each voxel is contaminated by the metal artifacts. The greater the likelihood that a voxel is contaminated by metal artifacts, the more the external energy is reduced by the weighting module 24. In this manner, the shape of the model becomes the predominant force where there is increased uncertainty regarding voxels of the diagnostic image.
- the techniques for locating the metal objects in the image also provide for removing the metal image and removing the artifacts from the image.
- the artifacts are removed by estimating or guessing which structure should replace the artifact.
- metal artifacts can be removed by deconstructing the image into sinograms, replacing data in the sinograms in the metal shadow by interpolated data, and reconstructing the image.
- the artifact-corrected images can contain uncertainties, although the uncertainties are distributed differently than in the artifact contaminated image.
- a selected organ model with a triangulated mesh surface, and a set of adaptation parameters is selected from the memory 20.
- the model generator receives a data set from the SID and/or the reconstruction processor, (e.g., a CT image, MR image, 3D anatomical image, etc.)
- N ⁇ is the number of triangles in the mesh surface
- w t is the
- ⁇ t is the coordinates of the best feature for triangle t
- ⁇ t is the coordinates of the center of triangle t .
- FIGURE 2 illustrates an image of a hip with two implants, and an uncertainty map such as may be generated by the voxel analyzer.
- an uncertainty measure is derived, which is passed forward to the GUI and/or to the segmentation tool.
- Watzke and208 e.g. Watzke and208
- That method involves metal identification and removal of metal artifacts, which in turn includes forward projection of the metal, superlinear weighting (e.g. cubic), and unweighted back-projection.
- the resulting image is represented by the uncertainty map 42.
- the structure of the image in map 42 correlates well with the appearance of metal artifacts in the original image 40.
- the uncertainty map can now be used to balance locally the weight of the image term and the model term in a model based segmentation of the image 40.
- FIGURE 3 shows an example of the technique as applied to the example of
- Figure 2 together with a metal artifact reduction method (e.g., such as is described in Koehler, "Apparatus, method, and computer program for producing a corrected image of a region of interest from acquired projection data," EP07100081, filed on January 4, 2007).
- the method involves an interpolation of projection data across the shadow of the metal.
- the uncertainty for the image 50 can be modeled for each voxel as the fraction of interpolated projection samples that contributed to the voxel.
- both images 40 and 50 original and corrected
- FIGURE 4 is a diagram illustrating some of the basic principles of model- based segmentation.
- Model-based segmentation includes a training phase 72 and a segmentation phase 74, which may occur sequentially and/or concurrently.
- training data 76 comprising a reference set of segmented images and corresponding mesh models is compiled, analyzed, and/or processed to generate one or more trained models 78.
- Each trained model has a mean mesh, shape variability, and identified features (e.g., contours, surface landmarks, etc.) at its boundaries.
- a new 3D image 80 is received (e.g., at the model generator 16, etc.), such as a reconstructed image (e.g., CT, PET, SPECT, MRI, fMRI, etc.) Positioning of the new image 80 relative to the trained model 78 is performed at 82.
- the new image is adapted to the model (e.g., by applying internal and external forces), and the segmented image 86 is generated.
- the segmented model may then be employed for one or more therapy planning or the like (e.g., radiation therapy, ablation therapy, surgery, etc.)
- FIGURE 5 is an example of segmented image adaptation of the image 80 and a trained model 78 using external forces 100 that push inward on the model to adapt the model to the image.
- the external force energy is characterized by the features of the dataset (e.g., gradients), organ-model-specific adaptation parameters (e.g., maximum gradient), the shape of the mesh (e.g., normal vector of triangles in the mesh), etc.
- a feature 102 is weak (e.g., in terms of external energy) since the features of the model and image are substantially parallel and aligned, and a reconfigured mesh does not intersect a best feature, thus the external energy applied to the feature 102 is small.
- Feature 104 has a large external energy contribution since the reconfigured mesh intersects the best feature of the image.
- Feature 106 has a moderate external energy contribution because the reconfigured mesh is near the best feature in this region.
- E ext external energy
- N ⁇ is the number of triangles in the mesh surface
- w t is the feature strength of a triangle t
- x t is the coordinates of the best feature for triangle t
- x t is the coordinates of the center of triangle t.
- FIGURE 6 shows features of the model 78 and their associations with features of the image 80, which delineate the internal energies of the respective features.
- the internal energy of the system is characterized by the shape of the mesh (e.g., the more the mesh deforms, the greater the increase in internal energy), the shape of the model of the anatomical structure, or mean model (e.g., when the mesh and the mean model have the same shape, the internal energy is zero), and the transformation (e.g., scaling factor and rotation vector) used to register the mean mesh and current configuration of the mesh.
- feature 104 has a large internal energy contribution, since the model and the image intersect each other.
- Feature 106 has a moderate internal energy contribution
- figure 108 has a low internal energy contribution.
- E int is internal energy
- N e d g es is the number of edges in the mesh
- ⁇ e is a vector representing edge e
- s is a mean mesh scaling factor
- R is the mean mesh rotation vector
- ⁇ ° and x € is a mean mesh vector representing edge e.
- FIGURE 7 illustrates an adaptation iteration wherein the mesh model is adapted to the image.
- model-based segmentation is an iterative algorithm.
- the system assigns a "best feature" and a feature weight to each triangle of the mesh.
- the external energies 100 and internal energies 120 are computed using the best features and weights.
- the external and internal energies are represented by a set of equations.
- the set of equations is solved (e.g., a matrix is inverted), and the solution of the system of equations is the new configuration (shape) of the mesh.
- arrows illustrate movement of mesh model nodes toward corresponding image nodes.
- FIGURE 8 illustrates an example of an internal force instruction in execution on a model 78, such as may be executed or prompted by the segmentation tool 26 in conjunction with a CT scanner or the like.
- an instruction such as "take 21 steps along the triangle normal vector" (e.g., from 10 mm inside the mesh to 10 mm outside the mesh) is executed.
- the interpolated CT value and the gradient vector are computed.
- the CT values and gradient vectors are sufficient information for the model-based segmenting system to operate.
- FIGURE 9 shows a screenshot 150 showing an organ surface 80 with a mesh model 78 surface overlaid thereon.
- a triangle normal vector 154 extends perpendicularly through a triangle of the mesh model surface. If internal or external energies warrant, the mesh model surface at this triangle will be moved along the triangle normal vector. In the example of Figure 9, external forces can push the mesh triangle through which the triangle normal vector passes toward the organ surface.
- a graph 158 of profile data along the triangle normal vector illustrates relative positions of pixels related to CT data points.
- Figures 10-11 illustrate one or more methods related to generating uncertainty maps and compensating for distortion and/or artifacts in patient images caused by metal in the patient, in accordance with various features. While the methods are described as a series of acts, it will be understood that not all acts may be required to achieve the described goals and/or outcomes, and that some acts may, in accordance with certain aspects, be performed in an order different that the specific orders described.
- FIGURE 10 illustrates a method 160 for generating uncertainty maps to mitigate incorrect segmentation results caused by metal objects and corresponding artifacts in a patient when segmenting a radiographic image of the patient.
- the likelihood that a voxel in a patient image is contaminated by a metal artifact is quantified.
- the quantification includes determining whether metal is present in an imaged region of the patient.
- a level of likelihood of metal artifact being present is qualitatively predicted.
- the qualitative prediction is performed using a method set forth in Watzke and208 (Eur Radiol 14:849-856, 2004), which describes a combination of linear interpolation of reprojected metal traces and multi-dimensional adaptive filtering of raw data in conjunction with directional weighting to reduce metal artifacts.
- Watzke and208 describes a combination of linear interpolation of reprojected metal traces and multi-dimensional adaptive filtering of raw data in conjunction with directional weighting to reduce metal artifacts.
- metal identification and artifact removal which includes forward projection, superlinear weighting (e.g. quadratic of cubic) followed by unfiltered back-projection.
- contaminated voxel data is replaced with interpolated data in an uncertainty map of the patient image.
- the updated model is then deformed to the image using model-based segmentation to segment the image, at 168.
- FIGURE 11 illustrates a method 180 of generating an uncertainty map to adapt a model of an anatomical structure for segmenting a patient image.
- projection data is interpolated across a shadowed area caused by a metal object in a patient to facilitate generating an uncertainty map.
- the interpolation is performed using a method described in Koehler, "Apparatus, method, and computer program for producing a corrected image of a region of interest from acquired projection data," EP07100081, filed on January 4, 2007.
- voxels are adjusted in the model to reflect the interpolated data.
- each voxel is adjusted proportionally to the fraction of interpolated projection samples that contribute to the voxel, relative to the original non-interpolated samples that contribute to the voxel.
- the adjusted model is applied to the patient image for model-based segmentation such as is described with regard to preceding figures.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Public Health (AREA)
- Radiology & Medical Imaging (AREA)
- Optics & Photonics (AREA)
- Heart & Thoracic Surgery (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Pathology (AREA)
- Veterinary Medicine (AREA)
- Biophysics (AREA)
- High Energy & Nuclear Physics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Geometry (AREA)
- Computer Graphics (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
When performing model-based segmentation on a 3D patient image (80), metal artifacts in the patient image (80), caused by metal in the patient's body, are detected, and a metal artifact reduction technique is performed to reduce the artifact(s) by interpolation projection data in the region of the artifact(s). The interpolated data is used to generate an uncertainty map for artifact- affected voxels in the image, and a mesh model (78) is conformed to the image to facilitate segmentation thereof. Internal and external energies applied to push and pull the model (78) are weighted as a function of the uncertainty associated with one or more voxels in the image (80). Iteratively, mathematical representations of the energies and respective weights are solved to describe an updated model shape that more closely aligns to the image (80).
Description
UNCERTAINTY MAPS FOR SEGMENTATION IN THE PRESENCE OF METAL
ARTIFACTS
DESCRIPTION
The present application finds particular application in model-based segmentation systems, particularly involving adaptive subject imaging. However, it will be appreciated that the described technique may also find application in other subject imaging systems, other image adaptation scenarios, or other medical techniques.
Metal implants such as hip replacements or dental fillings are a source of severe artifacts in CT imaging. A prominent and undesirable effect is that metal implants cause beam-hardening, which causes dark and bright shades close to and radiating from the metal. Metal artifacts impede proper diagnosis, as well as automatic or semi-automatic organ segmentation. Metal artifact reduction methods can improve subjective image quality for the observer. Regardless of the visual improvement, these methods are often based on removal of the artifacted voxels and replacement by an interpolation of nearby voxels, which introduces an additional uncertainty, such that that there is no guarantee that original structures are recovered by the method. Moreover, structures are often unaccounted for after metal artifact reduction, and other structures may be undesirably introduced. This imposes another impediment for automatic organ segmentation.
Image reconstruction and/or segmentation can be detrimentally affected by metal in a patient. In one example of model-based reconstruction, a mesh model of an organ is superimposed on a diagnostic image or diagnostic data. Two opposing forces, internal energy through which the model tries to hold its shape, and external energy that urges the model towards the interfaces or surfaces in the diagnostic image, work concurrently in opposition to bring the model to the shape of the organ. While model- based segmentation is satisfactory when the edges of the organ are well-defined in the image, problems can arise when there is artifact degradation, such as the metal artifacts discussed above. That is, metal objects (e.g., pins, screws, hip replacements, filings, etc.) in a patient create artifacts, such as streaks, which obscure image detail, increasing uncertainty regarding the exact location of the surface of the organ.
The present application provides new and improved model-based segmentation systems and methods that account for uncertainty associated with image data obscured by metal artifacts, which overcome the above-referenced problems and others.
In accordance with one aspect, a system for system for image segmentation in the presence of metal artifacts includes a model generator that receives patient image data and stores trained models of anatomical structures, and a voxel analyzer that determines whether metal artifacts are present in one or more voxels in the patient image data. The system further includes a processor that executes a metal artifact reduction algorithm and generates an uncertainty map with corrected voxel data incorporated therein for a patient image generated from the patient image data, and a segmentation tool that conforms a trained model of an anatomical structure corresponding to the patient image and segments the patient image using a model-based segmentation technique and evaluates the uncertainty derived by the processor. In accordance with another aspect, a method for performing model-based segmentation in includes generating a patient image of a region of a patient that includes a metal object, generating an uncertainty map for the patient image indicative of uncertainty in portions of the patient image due to metal object reconstruction artifacts, and employing the uncertainty map when segmenting a portion of the patient image displaced from the metal object using model-based segmenting.
In accordance with another aspect, an apparatus that facilitates model-based segmentation using uncertainty maps includes means for generating patient image data, means for reconstructing the patient image data into a 3D patient image, and means for detecting metal artifacts in one or more voxels in the 3D patient image. The apparatus further includes means for generating an uncertainty map with interpolated data that reduces the metal artifacts, means for weighting features associated with regions of a surface of a mesh model, and means for segmenting the 3D patient image using model- base segmentation to conform the mesh model to the 3D patient image.
In accordance with another aspect, a method of performing model-based segmentation comprises generating a diagnostic image of an organ or region of a subject, which image has metal artifacts, selecting a model of the imaged organ or region, and applying a first force on the model, which urges the model to hold its initial shape. The
method further comprises applying a second force on the model, which urges the model to deform into alignment with corresponding interfaces in the diagnostic image, and, on an image voxel by image voxel basis, adjusting the first and second forces in accordance with the degree of metal artifacting, such that the magnitude of the first force relative to the second force increases for voxels with a higher degree of metal artifacting, and the magnitude of the second force relative to the first force increases for voxels with a lower degree of metal artifacting.
Yet another aspect relates to a system for image segmentation in the presence of metal artifacts, including a model generator that receives patient image data and stores trained models of anatomical structures, a voxel analyzer that determines whether metal artifacts are present in one or more voxels in the patient image data, and a processor that generates an uncertainty map without changing the patient image data. The system further includes a segmentation tool that conforms a trained model of an anatomical structure corresponding to the patient image and segments the patient image using a model- based segmentation technique and evaluates the uncertainty derived by the processor.
One advantage is that image segmentation accuracy is improved.
Another advantage resides in mitigating metal artifact distortion in segmented images.
Another advantage resides in reducing total energy expended to deform a model to an image, which improves processing speed.
Still further advantages of the subject innovation will be appreciated by those of ordinary skill in the art upon reading and understand the following detailed description.
The innovation may take form in various components and arrangements of components, and in various steps and arrangements of steps. The drawings are only for purposes of illustrating various aspects and are not to be construed as limiting the invention.
FIGURE 1 illustrates a system that adjusts the external energy applied to deform an image in accordance with the likelihood that a voxel in the image is contaminated by metal artifacts.
FIGURE 2 illustrates an image of a hip with two implants, and an uncertainty map such as may be generated by the voxel analyzer.
FIGURE 3 shows an example of the technique as applied to the example of Figure 2, together with a metal artifact reduction method FIGURE 4 illustrates a diagram of the basic principles of model-based segmentation.
FIGURE 5 is an example of segmented image adaptation of the image and a trained model using external forces that push inward on the model to adapt the model to the image. FIGURE 6 shows features of the model and their associations with features of the image, which delineate the internal energies of the respective features.
FIGURE 7 illustrates an adaptation iteration wherein the mesh model is adapted to the image. It is to be appreciated that model-based segmentation is an iterative algorithm. FIGURE 8 illustrates an example of an internal force instruction in execution, such as may be executed or prompted by the segmentation tool in conjunction with a CT scanner or the like.
FIGURE 9 is an illustration of a screenshot showing an organ surface with a mesh model surface overlaid thereon. FIGURE 10 illustrates a method for generating uncertainty maps in to mitigate distortion caused by metal objects in a patient when segmenting a nuclear or radiographic image of the patient.
FIGURE 11 illustrates a method of generating an uncertainty map to adapt a model of an anatomical structure for segmenting a patient image.
The systems and methods described herein facilitate segmenting an image of soft tissue whose surface boundaries are artifacted and rendered uncertain by artifacts from a metal object in other regions of an imaged subject. For example, a metal ball in a hip replacement can cause a starburst-like artifact across an image of a patient's pelvic region, partially obscuring organs of interest (e.g., bladder, prostate, etc.). Not only does the metal project artifacts into remote areas of the image, the artifact obscures the precise
position of the metal itself. To improve image appearance, others have proposed trying to identify the exact location of the metal object, using this information to estimate a correction to the unreconstructed data, and reconstructing the corrected data. This and other techniques remove all or most of the starburst-like artifacts and display a reduced or substantially unartifacted image. However, due to the uncertainty associated with the identification of the true size and location of the metal object, and uncertainties in orther parts of the correction process, there remains uncertainty regarding the accuracy of the boundary surfaces of the remote organs in other parts of the image. The technique(s) described herein control or adjust for the uncertainty regarding the true location of the boundary surfaces of the remote organs during segmentation of these remote organs. In a first described embodiment, the remote organs are segmented by overlaying a mesh model having the shape of a normal organ. Opposing forces are applied to the mesh model. One force is applied to move the surface of the mesh model into alignment with surface boundaries in the image. The other force is applied to hold the mesh in the nominal organ shape. The relative magnitudes of these forces are adjusted in accordance with the uncertainty. It will be appreciated that this embodiment is applicable to uncertainties due to other factors, and does not require performing a metal artifact correction technique as a preliminary step.
Now turning to FIGURE 1, a system 10 is illustrated, which adjusts the external energy applied to deform an image in accordance with the likelihood that a voxel in the image is contaminated by metal artifacts. The system includes a subject imaging device (SID) 12 that is coupled to an image data memory 13 and a reconstruction processor 14, which receives captured data from the SID and reconstructs the data into an image. The SID can be, for instance, a computed tomography (CT) scanner, a positron emission tomography (PET) scanner, a single photon emission computed tomography (SPECT) scanner, a magnetic resonance imaging (MRI) device, a functional MRI (fMRI) device, or some other suitable subject imaging device. The system further includes a model generator 16, which comprises a processor 18 and memory 20. The memory can store information associated with one or more models of anatomical structures, which may be generated from one or more clinical images of subjects. For instance, the memory can store 3D models of individual organs or other structures (e.g., heart, liver, kidney, bladder, prostate, lung, salivary gland, etc.), as well as groups of organs and/or structures (e.g., liver-and-kidney,
heart-and-lung(s), pelvic region, abdominal region, thoracic region, cranial region, etc.). The stored models are models of the organ or structure for a nominal person and may further include models for the inter-patient variability of these organs. The model generator further includes a voxel analyzer 22 that quantitatively and qualitatively analyzes a subject image for artifacts caused by metal in the subject (e.g., fillings, hip replacements, pins, screws, etc.), and a weighting module 24 that weights the external energy locally at a voxel position according to the likelihood metal artifacts being present at this location. Furthermore, the model generator includes a segmentation tool 26 that applies external and internal forces to a selected model to conform the model to the subject image for segmentation while accounting for artifacts caused by metal objects. Additionally, the system comprises a graphical user interface (GUI) 28, which presents image and/or model information to a user and into which a user enters information to manipulate images, models, etc.
A metal identification protocol is applied by the processor 18, in conjunction with the SID 12, to identify the location of the metal. Once the metal is identified, the voxel analyzer 22 qualitatively predicts whether each voxel is contaminated by the metal artifacts. The greater the likelihood that a voxel is contaminated by metal artifacts, the more the external energy is reduced by the weighting module 24. In this manner, the shape of the model becomes the predominant force where there is increased uncertainty regarding voxels of the diagnostic image.
In another embodiment, the techniques for locating the metal objects in the image also provide for removing the metal image and removing the artifacts from the image. In this scenario, the artifacts are removed by estimating or guessing which structure should replace the artifact. For example, metal artifacts can be removed by deconstructing the image into sinograms, replacing data in the sinograms in the metal shadow by interpolated data, and reconstructing the image. Thus, the artifact-corrected images can contain uncertainties, although the uncertainties are distributed differently than in the artifact contaminated image.
According to an example, a selected organ model with a triangulated mesh surface, and a set of adaptation parameters is selected from the memory 20. When the model generator receives a data set from the SID and/or the reconstruction processor, (e.g., a CT image, MR image, 3D anatomical image, etc.), the processor executes an objective
function Etotai = wint x Eint + wext x Eext, wherein Etotai is total energy, Eint is internal energy, E6Xt is external energy, wint is an internal energy weight, wext is an external energy weight.
The external energy is Eext = ~S\ r(xt)wt {xt - xt ) where r(xt) represents the reliability of
the external energy or force (e.g., a fraction of original data contributing locally to the image) and is spatially variant, NΔ is the number of triangles in the mesh surface, wt is the
feature strength of a triangle t, χt is the coordinates of the best feature for triangle t, and χt is the coordinates of the center of triangle t .
In one embodiment, scaling the objective function (Etotai) does not change the location of the minimum (l/wext) x Etotai = (wint/weχt) x Eint + Eext. Therefore, wext can be set to 1 without constraining the problem, and the objective function becomes: Etotai =
Wlnt X Elnt + Eext.
FIGURE 2 illustrates an image of a hip with two implants, and an uncertainty map such as may be generated by the voxel analyzer. From the image 40, and/or as a byproduct of a metal artifact reduction method or technique, an uncertainty measure is derived, which is passed forward to the GUI and/or to the segmentation tool. If segmentation is done on the original images, a method by Watzke and Kalender (Eur Radiol 14:849-856, 2004), to estimate qualitatively the likelihood that a voxel suffers from beam-hardening, can be employed. That method involves metal identification and removal of metal artifacts, which in turn includes forward projection of the metal, superlinear weighting (e.g. cubic), and unweighted back-projection. The resulting image is represented by the uncertainty map 42. The structure of the image in map 42 correlates well with the appearance of metal artifacts in the original image 40. The uncertainty map can now be used to balance locally the weight of the image term and the model term in a model based segmentation of the image 40. FIGURE 3 shows an example of the technique as applied to the example of
Figure 2, together with a metal artifact reduction method (e.g., such as is described in Koehler, "Apparatus, method, and computer program for producing a corrected image of a region of interest from acquired projection data," EP07100081, filed on January 4, 2007). The method involves an interpolation of projection data across the shadow of the metal. Here, the uncertainty for the image 50 can be modeled for each voxel as the fraction of interpolated projection samples that contributed to the voxel. In a more advanced scenario,
both images 40 and 50 (original and corrected) can be used in the segmentation process together with the corresponding uncertainty maps 52. In addition to the image 50, which shows the original image after a metal artifact reduction method is applied, and the uncertainty map 52, image 54 shows uncertainties marked with iso-lines 56. FIGURE 4 is a diagram illustrating some of the basic principles of model- based segmentation. Model-based segmentation includes a training phase 72 and a segmentation phase 74, which may occur sequentially and/or concurrently. During the training phase, training data 76 comprising a reference set of segmented images and corresponding mesh models is compiled, analyzed, and/or processed to generate one or more trained models 78. Each trained model has a mean mesh, shape variability, and identified features (e.g., contours, surface landmarks, etc.) at its boundaries. During the segmentation phase, a new 3D image 80 is received (e.g., at the model generator 16, etc.), such as a reconstructed image (e.g., CT, PET, SPECT, MRI, fMRI, etc.) Positioning of the new image 80 relative to the trained model 78 is performed at 82. At 84, the new image is adapted to the model (e.g., by applying internal and external forces), and the segmented image 86 is generated. The segmented model may then be employed for one or more therapy planning or the like (e.g., radiation therapy, ablation therapy, surgery, etc.)
FIGURE 5 is an example of segmented image adaptation of the image 80 and a trained model 78 using external forces 100 that push inward on the model to adapt the model to the image. The external force energy is characterized by the features of the dataset (e.g., gradients), organ-model-specific adaptation parameters (e.g., maximum gradient), the shape of the mesh (e.g., normal vector of triangles in the mesh), etc.
As illustrated, a feature 102 is weak (e.g., in terms of external energy) since the features of the model and image are substantially parallel and aligned, and a reconfigured mesh does not intersect a best feature, thus the external energy applied to the feature 102 is small. Feature 104 has a large external energy contribution since the reconfigured mesh intersects the best feature of the image. Feature 106 has a moderate external energy contribution because the reconfigured mesh is near the best feature in this region.
'=NA r ~ Λ Λ2 External energy is determined by the equation: Eext = ∑ r(xt)wt\ xt- x, ,
where Eext is external energy, NΔ is the number of triangles in the mesh surface, wt is the
feature strength of a triangle t, xt is the coordinates of the best feature for triangle t, and xt is the coordinates of the center of triangle t.
FIGURE 6 shows features of the model 78 and their associations with features of the image 80, which delineate the internal energies of the respective features. The internal energy of the system is characterized by the shape of the mesh (e.g., the more the mesh deforms, the greater the increase in internal energy), the shape of the model of the anatomical structure, or mean model (e.g., when the mesh and the mean model have the same shape, the internal energy is zero), and the transformation (e.g., scaling factor and rotation vector) used to register the mean mesh and current configuration of the mesh. In the figure, feature 104 has a large internal energy contribution, since the model and the image intersect each other. Feature 106 has a moderate internal energy contribution, and figure 108 has a low internal energy contribution.
wherein Eint is internal energy, Nedges is the number of edges in the mesh, χe is a vector representing edge e, s is a mean mesh scaling factor, R is the mean mesh rotation vector,
→° and x€ is a mean mesh vector representing edge e.
FIGURE 7 illustrates an adaptation iteration wherein the mesh model is adapted to the image. It is to be appreciated that model-based segmentation is an iterative algorithm. At each iteration, the system assigns a "best feature" and a feature weight to each triangle of the mesh. The external energies 100 and internal energies 120 are computed using the best features and weights. Mathematically, the external and internal energies are represented by a set of equations. The set of equations is solved (e.g., a matrix is inverted), and the solution of the system of equations is the new configuration (shape) of the mesh. In Figure 7, arrows illustrate movement of mesh model nodes toward corresponding image nodes.
FIGURE 8 illustrates an example of an internal force instruction in execution on a model 78, such as may be executed or prompted by the segmentation tool 26 in conjunction with a CT scanner or the like. According to the example, an instruction such as "take 21 steps along the triangle normal vector" (e.g., from 10 mm inside the mesh
to 10 mm outside the mesh) is executed. At each step, the interpolated CT value and the gradient vector are computed. The CT values and gradient vectors are sufficient information for the model-based segmenting system to operate.
FIGURE 9 shows a screenshot 150 showing an organ surface 80 with a mesh model 78 surface overlaid thereon. A triangle normal vector 154 extends perpendicularly through a triangle of the mesh model surface. If internal or external energies warrant, the mesh model surface at this triangle will be moved along the triangle normal vector. In the example of Figure 9, external forces can push the mesh triangle through which the triangle normal vector passes toward the organ surface. A graph 158 of profile data along the triangle normal vector illustrates relative positions of pixels related to CT data points.
Figures 10-11 illustrate one or more methods related to generating uncertainty maps and compensating for distortion and/or artifacts in patient images caused by metal in the patient, in accordance with various features. While the methods are described as a series of acts, it will be understood that not all acts may be required to achieve the described goals and/or outcomes, and that some acts may, in accordance with certain aspects, be performed in an order different that the specific orders described.
FIGURE 10 illustrates a method 160 for generating uncertainty maps to mitigate incorrect segmentation results caused by metal objects and corresponding artifacts in a patient when segmenting a radiographic image of the patient. At 162, the likelihood that a voxel in a patient image is contaminated by a metal artifact is quantified. In one embodiment, the quantification includes determining whether metal is present in an imaged region of the patient. At 164, a level of likelihood of metal artifact being present (e.g., beam-hardening) is qualitatively predicted. In one embodiment, the qualitative prediction is performed using a method set forth in Watzke and Kalender (Eur Radiol 14:849-856, 2004), which describes a combination of linear interpolation of reprojected metal traces and multi-dimensional adaptive filtering of raw data in conjunction with directional weighting to reduce metal artifacts. In this approach, the likelihood of metal artifacts appearance due to beam-hardening is estimated by metal identification and artifact removal, which includes forward projection, superlinear weighting (e.g. quadratic of cubic) followed by unfiltered back-projection.
At 166, contaminated voxel data is replaced with interpolated data in an uncertainty map of the patient image. The updated model is then deformed to the image using model-based segmentation to segment the image, at 168.
FIGURE 11 illustrates a method 180 of generating an uncertainty map to adapt a model of an anatomical structure for segmenting a patient image. At 182, projection data is interpolated across a shadowed area caused by a metal object in a patient to facilitate generating an uncertainty map. In one embodiment, the interpolation is performed using a method described in Koehler, "Apparatus, method, and computer program for producing a corrected image of a region of interest from acquired projection data," EP07100081, filed on January 4, 2007. At 184, voxels are adjusted in the model to reflect the interpolated data. For instance, each voxel is adjusted proportionally to the fraction of interpolated projection samples that contribute to the voxel, relative to the original non-interpolated samples that contribute to the voxel. At 186, the adjusted model is applied to the patient image for model-based segmentation such as is described with regard to preceding figures.
Claims
1. A system (10) for image segmentation in the presence of metal artifacts, including: a model generator (16) that receives patient image data and stores trained models of anatomical structures; a voxel analyzer (22) that determines whether metal artifacts are present in one or more voxels in the patient image data; a processor (18) that executes a metal artifact reduction algorithm and generates an uncertainty map with corrected voxel data incorporated therein for a patient image (80) generated from the patient image data; and a segmentation tool (26) that conforms a trained model (78) of an anatomical structure corresponding to the patient image and segments the patient image using a model- based segmentation technique and evaluates the uncertainty derived by the processor (18).
2. The system according to claim 1 , wherein the model (78) has a triangulated mesh surface.
3. The system of claim 1, wherein a feature on the surface of the model (78) is selected at least one of automatically or in response to user input.
4. The system according to claim 3, wherein the segmentation tool (26) applies an internal force along a surface normal vector in a surface region in which the feature is located.
5. The system according to claim 4, wherein the segmentation tool (26) applies an external force along the vector.
6. The system according to claim 5, wherein the segmentation tool (26) balances the internal force and the external force to align the surface region to a surface of the patient image (80).
7. The system according to claim 5, wherein the total energy applied to the feature through the internal and external forces is expressed as Etotai = wint x Eint + wext x E6Xt, wherein Etotai is total energy, Eint is internal energy, Eext is external energy, wint is an internal energy weight, and wext is an external energy weight, and wherein
Eext = y^ r(xt)wt(xt - X, )2, where r(xt) represents the reliability of the external energy
(=1 and is spatially variant, NΔ is the number of triangles in the mesh surface, wt is the feature
strength of a triangle t, xt is the coordinates of the best feature for triangle t, and xt is the coordinates of the center of triangle t.
8. The system according to claim 7, wherein wext is 1, and wherein the segmentation tool (26) applies increased internal force to compensate for a detected metal artifact.
9. The system according to claim 1, wherein the model generator (16) includes: a routine (162) or means for determining whether a voxel is affected by a metal artifact; a routine (164) or means for quantitatively predicting a level to which the voxel is affected by the metal artifact; a routine (166) or means for interpolating projection data and updating the voxel with the interpolated projection data; and a routine (168) or means for deforming the model (78) to the patient image (80).
10. A method of performing model-based segmentation in the system (10) of claim 1, including: determining whether a voxel is affected by a metal artifact; quantitatively predicting a level to which the voxel is affected by the metal artifact; interpolating projection data and updating the voxel with the interpolated projection data; and deforming the model (78) to the patient image (80).
11. A method for performing model-based segmentation, including: generating a patient image (80) of a region of a patient that includes a metal object; generating an uncertainty map for the patient image (80) indicative of uncertainty in portions of the patient image due to metal object reconstruction artifacts; and employing the uncertainty map when segmenting a portion of the patient image (80) displaced from the metal object using model-based segmenting.
12. The method according to claim 11, further including overlaying a mesh model (78) on the patient image (80) and aligning corresponding features in the model (78) and the patient image (80).
13. The method according to claim 12, further including applying external forces to the mesh model (78) to pull one or more features on the surface of the mesh model (78) inward toward corresponding image features.
14. The method according to claim 13, further including applying internal forces to the mesh model surface to pull one or more features outward toward corresponding image features.
15. The method according to claim 14, further comprising weighting the internal forces as a function of a degree to which voxels in the one or more features are affected by the metal.
16. The method according to claim 12, wherein model-based segmenting includes: assigning a best feature and a feature weight to each triangle of a plurality of triangles that comprise the surface of the mesh model (78); computing external and internal energies to be applied to each triangle using the respective best features and weights; modeling the external an internal energies as mathematical equations; solving the mathematical equations; and updating the shape of the mesh model (78) as a function of the equation solutions.
17. The method according to claim 16, further including modeling the external energy as:
Ee* = ∑ r<Λ)w((x( - x()2,
wherein Eext is external energy, NΔ is the number of triangles in the mesh surface, r(xt) represents the reliability of the external energy and is spatially variant, wt is
the feature strength of a triangle t, X1 is the coordinates of the best feature for triangle t, and
jet is the coordinates of the center of triangle t.
18. The method according to claim 16, further including modeling the internal energy as:
wherein Eint is internal energy, Nedges is the number of edges in the mesh, xe is a vector representing edge e, s is a mean mesh scaling factor, R is the mean mesh
→° rotation vector, and xc is a mean mesh vector representing edge e.
19. The method according to claim 16, performed iteratively until the mesh model (78) conforms to the patient image (80).
20. A processor (18) or computer medium (20) programmed to perform the method of claim 11.
21. The method of claim 11, further including reducing metal artifacts in the patient image (80).
22. A system that facilitates model-based segmentation using uncertainty maps, including: means (12) for generating patient image data; means (14) for reconstructing the patient image data into a 3D patient image (80). means (22) for detecting metal artifacts in one or more voxels in the 3D patient image (80); means (16) for generating an uncertainty map with interpolated data that reduces the metal artifacts; means (24) for weighting features associated with regions of a surface of a mesh model (78); and means (26) for segmenting the 3D patient image (80) using model-base segmentation to conform the mesh model (78) to the 3D patient image (80).
23. A method of performing model-based segmentation, comprising: generating a diagnostic image of an organ or region of a subject, which image has metal artifacts; selecting a model of the imaged organ or region; applying a first force on the model, which urges the model to hold its initial shape; applying a second force on the model, which urges the model to deform into alignment with corresponding interfaces in the diagnostic image; on an image voxel by image voxel basis, adjusting the first and second forces in accordance with the degree of metal artifacting, such that the magnitude of the first force relative to the second force increases for voxels with a higher degree of metal artifacting, and the magnitude of the second force relative to the first force increases for voxels with a lower degree of metal artifacting.
24. A system (10) for image segmentation in the presence of metal artifacts, including: a model generator (16) that receives patient image data and stores trained models of anatomical structures; a voxel analyzer (22) that determines whether metal artifacts are present in one or more voxels in the patient image data; a processor (18) that generates an uncertainty map without changing the patient image data; and a segmentation tool (26) that conforms a trained model (78) of an anatomical structure corresponding to the patient image and segments the patient image using a model- based segmentation technique and evaluates the uncertainty derived by the processor (18).
Priority Applications (5)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US12/673,779 US9730663B2 (en) | 2007-08-31 | 2008-08-12 | Uncertainty maps for segmentation in the presence of metal artifacts |
CN201611070118.7A CN106920246B (en) | 2007-08-31 | 2008-08-12 | Uncertainty map for segmentation in the presence of metal artifacts |
EP08807295A EP2188776B1 (en) | 2007-08-31 | 2008-08-12 | Uncertainty maps for segmentation in the presence of metal artifacts |
AT08807295T ATE511161T1 (en) | 2007-08-31 | 2008-08-12 | UNCERTAINTY MAP FOR SEGMENTATION IN THE PRESENCE OF METAL ARTIFACTS |
CN200880104577XA CN102067170A (en) | 2007-08-31 | 2008-08-12 | Uncertainty maps for segmentation in the presence of metal artifacts |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US96917607P | 2007-08-31 | 2007-08-31 | |
US60/969,176 | 2007-08-31 |
Publications (2)
Publication Number | Publication Date |
---|---|
WO2009027889A2 true WO2009027889A2 (en) | 2009-03-05 |
WO2009027889A3 WO2009027889A3 (en) | 2010-03-25 |
Family
ID=40387959
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/IB2008/053236 WO2009027889A2 (en) | 2007-08-31 | 2008-08-12 | Uncertainty maps for segmentation in the presence of metal artifacts |
Country Status (5)
Country | Link |
---|---|
US (1) | US9730663B2 (en) |
EP (1) | EP2188776B1 (en) |
CN (2) | CN106920246B (en) |
AT (1) | ATE511161T1 (en) |
WO (1) | WO2009027889A2 (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE102010034918A1 (en) * | 2010-08-20 | 2012-02-23 | Siemens Aktiengesellschaft | Method and apparatus for providing quality information for X-ray imaging |
WO2015116994A1 (en) | 2014-02-03 | 2015-08-06 | Vanderbilt University | Methods for automatic segmentation of inner ear anatomy in post-implantation ct and applications of same |
CN105843212A (en) * | 2016-03-29 | 2016-08-10 | 东北大学 | System and method for fault diagnosis of blast furnace |
US10460448B2 (en) | 2012-02-14 | 2019-10-29 | Koninklijke Philips N.V. | Method for quantification of uncertainty of contours in manual and auto segmenting algorithms |
CN112884699A (en) * | 2019-11-13 | 2021-06-01 | 西门子医疗有限公司 | Method and image processing device for segmenting image data and computer program product |
Families Citing this family (32)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8073252B2 (en) * | 2006-06-09 | 2011-12-06 | Siemens Corporation | Sparse volume segmentation for 3D scans |
JP6007102B2 (en) * | 2009-06-24 | 2016-10-12 | コーニンクレッカ フィリップス エヌ ヴェKoninklijke Philips N.V. | Determination of structural contours based on image information |
US8611670B2 (en) * | 2010-02-25 | 2013-12-17 | The Board Of Trustees Of The Leland Stanford Junior University | Intelligent part identification for use with scene characterization or motion capture |
US9763587B2 (en) | 2010-06-10 | 2017-09-19 | Biosense Webster (Israel), Ltd. | Operator-controlled map point density |
EP2754122B1 (en) * | 2011-11-08 | 2019-01-23 | Koninklijke Philips N.V. | Adaptive application of metal artifact correction algorithms |
US8891847B2 (en) * | 2012-01-23 | 2014-11-18 | Medtronic Navigation, Inc. | Automatic implant detection from image artifacts |
JP6243535B2 (en) * | 2013-12-10 | 2017-12-06 | コーニンクレッカ フィリップス エヌ ヴェKoninklijke Philips N.V. | Model-based segmentation of anatomical structures |
US10111714B2 (en) * | 2014-01-27 | 2018-10-30 | Align Technology, Inc. | Adhesive objects for improving image registration of intraoral images |
KR102220226B1 (en) * | 2014-02-12 | 2021-02-25 | 삼성전자주식회사 | Computer tomography apparatus and method for reconstructing a computer tomography image thereof |
US8958623B1 (en) | 2014-04-29 | 2015-02-17 | Heartflow, Inc. | Systems and methods for correction of artificial deformation in anatomic modeling |
JP6375575B2 (en) * | 2014-10-22 | 2018-08-22 | 朝日レントゲン工業株式会社 | Image processing apparatus, image processing method, and X-ray imaging apparatus |
US9704256B2 (en) | 2015-03-10 | 2017-07-11 | Siemens Healthcare Gmbh | Systems and method for computation and visualization of segmentation uncertainty in medical images |
CN107850649B (en) | 2015-06-26 | 2020-11-03 | 皇家飞利浦有限公司 | Magnetic resonance imaging system and method |
WO2017045618A1 (en) * | 2015-09-15 | 2017-03-23 | Shanghai United Imaging Healthcare Co., Ltd. | Image reconstruction system and method |
US9697623B1 (en) * | 2015-09-15 | 2017-07-04 | Shanghai United Imaging Healthcare Co., Ltd. | Image reconstruction system and method |
JP6780948B2 (en) * | 2016-03-28 | 2020-11-04 | キヤノンメディカルシステムズ株式会社 | X-ray CT device |
IL245339A (en) * | 2016-04-21 | 2017-10-31 | Rani Ben Yishai | Method and system for registration verification |
JP7256115B2 (en) | 2016-08-22 | 2023-04-11 | コーニンクレッカ フィリップス エヌ ヴェ | Model-normalized motion-compensated medical image reconstruction |
US9905044B1 (en) * | 2016-08-25 | 2018-02-27 | General Electric Company | Systems and methods for functional imaging |
US11138790B2 (en) | 2016-10-14 | 2021-10-05 | Axial Medical Printing Limited | Method for generating a 3D physical model of a patient specific anatomic feature from 2D medical images |
GB201617507D0 (en) | 2016-10-14 | 2016-11-30 | Axial3D Limited | Axial3D UK |
EP3646288B1 (en) * | 2017-06-26 | 2022-09-07 | Elekta, Inc. | Method for improving cone-beam ct image quality using a deep convolutional neural network |
KR102581685B1 (en) | 2017-06-30 | 2023-09-25 | 프로메이톤 홀딩 비.브이. | Classification and 3D modeling of 3D oral and maxillofacial structures using deep learning |
EP3462373A1 (en) | 2017-10-02 | 2019-04-03 | Promaton Holding B.V. | Automated classification and taxonomy of 3d teeth data using deep learning methods |
US11120551B2 (en) | 2017-11-27 | 2021-09-14 | Rensselaer Polytechnic Institute | Training a CNN with pseudo ground truth for CT artifact reduction |
EP3503038A1 (en) | 2017-12-22 | 2019-06-26 | Promaton Holding B.V. | Automated 3d root shape prediction using deep learning methods |
EP3561778A1 (en) * | 2018-04-26 | 2019-10-30 | Promaton Holding B.V. | Automated correction of metal affected voxel representations of x-ray data using deep learning techniques |
US11763502B2 (en) * | 2018-08-06 | 2023-09-19 | Vanderbilt University | Deep-learning-based method for metal reduction in CT images and applications of same |
US11309072B2 (en) | 2020-04-21 | 2022-04-19 | GE Precision Healthcare LLC | Systems and methods for functional imaging |
CN111862258A (en) * | 2020-07-23 | 2020-10-30 | 深圳高性能医疗器械国家研究院有限公司 | Image metal artifact suppression method |
GB202101908D0 (en) | 2021-02-11 | 2021-03-31 | Axial Medical Printing Ltd | Axial3D pathology |
CN116228916B (en) * | 2023-05-10 | 2023-07-11 | 中日友好医院(中日友好临床医学研究所) | Image metal artifact removal method, system and equipment |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6507633B1 (en) * | 2001-02-15 | 2003-01-14 | The Regents Of The University Of Michigan | Method for statistically reconstructing a polyenergetic X-ray computed tomography image and image reconstructor apparatus utilizing the method |
Family Cites Families (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6047080A (en) * | 1996-06-19 | 2000-04-04 | Arch Development Corporation | Method and apparatus for three-dimensional reconstruction of coronary vessels from angiographic images |
US6631212B1 (en) * | 1999-09-13 | 2003-10-07 | Eastman Kodak Company | Twostage scheme for texture segmentation based on clustering using a first set of features and refinement using a second set of features |
US6721387B1 (en) | 2001-06-13 | 2004-04-13 | Analogic Corporation | Method of and system for reducing metal artifacts in images generated by x-ray scanning devices |
DE10357206B4 (en) * | 2003-12-08 | 2005-11-03 | Siemens Ag | Method and image processing system for the segmentation of sectional image data |
JP2007520300A (en) * | 2004-02-05 | 2007-07-26 | コーニンクレッカ フィリップス エレクトロニクス エヌ ヴィ | Reduction of overall image artifacts caused by highly attenuated objects in CT distributing voxel tissue classes |
WO2005076222A1 (en) | 2004-02-09 | 2005-08-18 | Philips Intellectual Property & Standards Gmbh | Correction of artifacts caused by high attenuating objects |
DE102004022332A1 (en) * | 2004-05-06 | 2005-12-08 | Siemens Ag | Method for the post-reconstructive correction of images of a computer tomograph |
EP1603076A1 (en) * | 2004-05-13 | 2005-12-07 | Aalborg Universitet | Computerized cortex boundary extraction from MR images |
US7369695B2 (en) * | 2004-08-20 | 2008-05-06 | General Electric Company | Method and apparatus for metal artifact reduction in 3D X-ray image reconstruction using artifact spatial information |
WO2006039809A1 (en) | 2004-10-12 | 2006-04-20 | UNIVERSITé LAVAL | Method and apparatus for metal artifact reduction in computed tomography |
US8280136B2 (en) * | 2005-09-16 | 2012-10-02 | The Ohio State University | Method and apparatus for detecting intraventricular dyssynchrony |
US7711406B2 (en) * | 2005-11-23 | 2010-05-04 | General Electric Company | System and method for detection of electromagnetic radiation by amorphous silicon x-ray detector for metal detection in x-ray imaging |
US20080095300A1 (en) * | 2006-10-05 | 2008-04-24 | General Electric Company | System and method for iterative reconstruction using parallel processing |
US20080085040A1 (en) * | 2006-10-05 | 2008-04-10 | General Electric Company | System and method for iterative reconstruction using mask images |
CN101573731B (en) | 2007-01-04 | 2015-07-22 | 皇家飞利浦电子股份有限公司 | Apparatus and method and for producing a corrected image of a region of interest from acquired projection data |
US8311287B2 (en) * | 2008-09-25 | 2012-11-13 | Microsoft Corporation | Validation and correction of map data using oblique images |
-
2008
- 2008-08-12 WO PCT/IB2008/053236 patent/WO2009027889A2/en active Application Filing
- 2008-08-12 EP EP08807295A patent/EP2188776B1/en active Active
- 2008-08-12 AT AT08807295T patent/ATE511161T1/en not_active IP Right Cessation
- 2008-08-12 CN CN201611070118.7A patent/CN106920246B/en active Active
- 2008-08-12 CN CN200880104577XA patent/CN102067170A/en active Pending
- 2008-08-12 US US12/673,779 patent/US9730663B2/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6507633B1 (en) * | 2001-02-15 | 2003-01-14 | The Regents Of The University Of Michigan | Method for statistically reconstructing a polyenergetic X-ray computed tomography image and image reconstructor apparatus utilizing the method |
Non-Patent Citations (3)
Title |
---|
GU J ET AL: "Metal artifacts reduction in CT images through Euler's elastica and curvature based sinogram inpainting" PROCEEDINGS OF THE SPIE - THE INTERNATIONAL SOCIETY FOR OPTICAL ENGINEERING, SPIE, US, vol. 6144 III, 13 February 2006 (2006-02-13), pages 614465-1, XP002518900 ISSN: 0277-786X * |
WEESE J ET AL: "SHAPE CONSTRAINED DEFORMABLE MODELS FOR 3D MEDICAL IMAGE SEGMENTATION" LECTURE NOTES IN COMPUTER SCIENCE, SPRINGER VERLAG, DE, vol. 2082, 18 June 2001 (2001-06-18), pages 380-387, XP009027152 ISSN: 0302-9743 * |
ZHANG ET AL: "Reducing metal artifacts in cone-beam CT images by preprocessing projection data" INTERNATIONAL JOURNAL OF RADIATION: ONCOLOGY BIOLOGY PHYSICS, PERGAMON PRESS, USA, vol. 67, no. 3, 9 February 2007 (2007-02-09), pages 924-932, XP005897014 ISSN: 0360-3016 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE102010034918A1 (en) * | 2010-08-20 | 2012-02-23 | Siemens Aktiengesellschaft | Method and apparatus for providing quality information for X-ray imaging |
US10460448B2 (en) | 2012-02-14 | 2019-10-29 | Koninklijke Philips N.V. | Method for quantification of uncertainty of contours in manual and auto segmenting algorithms |
WO2015116994A1 (en) | 2014-02-03 | 2015-08-06 | Vanderbilt University | Methods for automatic segmentation of inner ear anatomy in post-implantation ct and applications of same |
EP3102110A4 (en) * | 2014-02-03 | 2019-03-27 | Vanderbilt University | Methods for automatic segmentation of inner ear anatomy in post-implantation ct and applications of same |
CN105843212A (en) * | 2016-03-29 | 2016-08-10 | 东北大学 | System and method for fault diagnosis of blast furnace |
CN112884699A (en) * | 2019-11-13 | 2021-06-01 | 西门子医疗有限公司 | Method and image processing device for segmenting image data and computer program product |
CN112884699B (en) * | 2019-11-13 | 2024-03-12 | 西门子医疗有限公司 | Method and image processing device for segmenting image data and computer program product |
Also Published As
Publication number | Publication date |
---|---|
WO2009027889A3 (en) | 2010-03-25 |
ATE511161T1 (en) | 2011-06-15 |
US9730663B2 (en) | 2017-08-15 |
CN106920246B (en) | 2024-03-08 |
US20110038516A1 (en) | 2011-02-17 |
EP2188776B1 (en) | 2011-05-25 |
EP2188776A2 (en) | 2010-05-26 |
CN106920246A (en) | 2017-07-04 |
CN102067170A (en) | 2011-05-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US9730663B2 (en) | Uncertainty maps for segmentation in the presence of metal artifacts | |
Mattes et al. | PET-CT image registration in the chest using free-form deformations | |
Fung et al. | Cerebral blood flow with [15O] water PET studies using an image-derived input function and MR-defined carotid centerlines | |
US8265356B2 (en) | Method and apparatus for efficient automated re-contouring of four-dimensional medical imagery using surface displacement fields | |
US9349198B2 (en) | Robust artifact reduction in image reconstruction | |
US9659364B2 (en) | Probabilistic refinement of model-based segmentation | |
Wallach et al. | Super-resolution in respiratory synchronized positron emission tomography | |
EP3128919B1 (en) | Lung segmentation and bone suppression techniques for radiographic images | |
US10388037B2 (en) | Selection method for an artifact correction algorithm, data processing facility for performing the method, and medical imaging system | |
Otte | Elastic registration of fMRI data using Bezier-spline transformations | |
JP2014509037A (en) | Processing based on image data model | |
Baka et al. | Statistical shape model-based femur kinematics from biplane fluoroscopy | |
WO2007057845A1 (en) | Method for delineation of predetermined structures in 3d images | |
Schmid et al. | Segmentation of X-ray images by 3D-2D registration based on multibody physics | |
EP3658031B1 (en) | Motion compensated cardiac valve reconstruction | |
Shieh et al. | Improving thoracic four-dimensional cone-beam CT reconstruction with anatomical-adaptive image regularization (AAIR) | |
JP2011516957A (en) | Mesh collision avoidance | |
US8284196B2 (en) | Method and system for reconstructing a model of an object | |
Rueckert et al. | Multiscale approach to contour fitting for MR images | |
Chica et al. | Example-guided segmentation | |
Ecabert et al. | Automatic whole heart segmentation in CT images: method and validation | |
US20230289933A1 (en) | System and method for time-series imaging | |
Gorthi et al. | Active contour-based segmentation of head and neck with adaptive atlas selection | |
US20240112331A1 (en) | Medical Image Data Processing Technique | |
Antila | Volumetric Image Segmentation for Planning of Therapies: Application to dental implants and muscle tumors |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 200880104577.X Country of ref document: CN |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 08807295 Country of ref document: EP Kind code of ref document: A2 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 12673779 Country of ref document: US |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2008807295 Country of ref document: EP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |